Reiser5 Pursuing Selective File Migration For Moving Hot Files To High Performance Disks

[starshipeleven]

Hey, here's an idea. How about renaming the ReiserFS so that it can have more than a snowflake's chance in hell of getting merged into Linus's kernel tree? Just a thought.

the name is not the issue, Linux kernel has not been overrun by SJWs. The only thing they care about is what actually matters.

[starshipeleven]

The name might not be an issue to upstream distributions since they can hide that behind their fancy desktop UI. Perhaps it's the Linux example of NIH - "not invented here". Look at Red Hat; if it ain't woven or weaveable into systemd and their other world domination efforts, then Red Hat (now IBM) does not want it. Same with Ubuntu. Not sure about SuSE.

To be fair, they are commercial entities, they need to make themselves different to sell more than a competitor

[reavertm]

Michael, you wrote "over the controversial Reiser4 file-system". Could you elaborate on what is controversial about Reiser4 specifically? Is in known for some spectacular data losses? Or when printed on A4 pages, first paragraph letters of source code are the same as in LaVey's Satanic Bible?
I genuinely would like to know.

[cjcox]

Chris lives in Texas. Chris is a male. Chris is Caucasian. Texas borders with Mexico. People cross the Mexico border into Texas. Some of those do so without legal rights to do so. Some illegal border crossers are a part of the drug trade, and many others are sold into sex based slavery/trafficking. Chris is against sex trafficking. Some of Chris's friends use drugs.

Conclusion: Chris is a drug dealing while supremist male chauvinist homosexual who hates all Mexicans. Alternatively, he might be a congressman.

(This was posted as humor, but feel free to pull if it causes riots, or whatever)

[waxhead]

That was a later (not so great) idea.

I found 3 attempts from IBM to get patches reviewed and then they turned silent.

I don't quite see why this would not work , but on the other hand I see benefits from not doing it as a general VFS thing.
For example BTRFS could theoretically reserve some space as a cache configured as RAID0 across the fastest devices on the pool and if the RAID0 stripe fails (or too many devices are busy) it could always fall back to RAID1 or whatever profile the original data is stored in. Of course this approach only works for reads, I have no clue how Reiserfs does it , but I assume it would be something similar.

[starshipeleven]

I don't quite see why this would not work , but on the other hand I see benefits from not doing it as a general VFS thing.
For example BTRFS could theoretically reserve some space as a cache configured as RAID0 across the fastest devices on the pool and if the RAID0 stripe fails (or too many devices are busy) it could always fall back to RAID1 or whatever profile the original data is stored in. Of course this approach only works for reads, I have no clue how Reiserfs does it , but I assume it would be something similar.

no you did not understand the point.

This is about the filesystem detecting what drives perform faster and keep track of what data is most often used. Then it decides to move this data to the faster drives while still maintainig the same RAID level.

What you say with a RAID0 cache is still the filesystem not aware of anything, and doing a dumb cache.

Afaik ZFS also has only half this feature, it can detect what data is "hot" but you need to designate cache drives (usually SSDs) for it.

[waxhead]

no you did not understand the point.

This is about the filesystem detecting what drives perform faster and keep track of what data is most often used. Then it decides to move this data to the faster drives while still maintainig the same RAID level.

Ah ok - thanks, so Reiserfs is keeping the same raid level but migrates the hot data to the fastest set of drives that can still satisfy the redundancy requirement. I was not even aware that reiserfs had RAID like functionality built in.

What you say with a RAID0 cache is still the filesystem not aware of anything, and doing a dumb cache.

Afaik ZFS also has only half this feature, it can detect what data is "hot" but you need to designate cache drives (usually SSDs) for it.

I would not agree that what I said about RAID0 would cause BTRFS to not be aware of anything, when it (if this was supported) could be aware of everything. I am not familiar enough with ZFS to comment on that except that I also think that cache drives needs to be designated for it.

To make such a cache automated you would actually need to track quite a few things. What data is hot, what drives are hot when accessing the hot data and more importantly what drives are cold as this would allow to make statistically smart choices at least in theory. In practice however I think that a designated set of drives is probably the most sane way to go.

[starshipeleven]

Ah ok - thanks, so Reiserfs is keeping the same raid level but migrates the hot data to the fastest set of drives that can still satisfy the redundancy requirement. I was not even aware that reiserfs had RAID like functionality built in.

This is Reiser 5, a "next-gen-filesystem" set of features added on top of Reiser 4 filesystem. It's in development.

[Awesomeness]

So you're telling me that the only thing stopping you from using Reiser4/5 is the name? I don't believe that.

I never said that. I was talking about the discussions. Nothing more, nothing less. Not really that hard to understand.